Skip to content

release: apis c9f3a0c to production - #309

Merged
jm-rivera merged 7 commits into
mainfrom
release/2026-11-14-3
Nov 14, 2026
Merged

jm-rivera merged 7 commits into
mainfrom
release/2026-11-14-3

Conversation

@jm-rivera

Copy link
Copy Markdown
Contributor

Releases apis at c9f3a0c to production, cut on release/2026-11-14-3: three pull requests after the release at 2f3b925.

Release order: release this before the SDK 0.2.0 tags. The SDK reads kept_in and the per-root reports when they are served and reads their absence as empty, so either order is safe. No migration, setting or secret change.

Testing: CI, Migrations and Deploy dev passed on c9f3a0c. On dev, the SDK journeys passed in both clients against #303, and "OOF to India by donor" returned 188 rows with 1968's DACOfficialDonors entry and its kept_in series. The SDK release preflight passes against production.

…2 recorded at 133262c

Moves all nine spec-kit pins to 7ef7f71. Its client contract adds the
timeout-and-retries rule, the dfo cache root with legacy readerkit
reads, the fetched/ cache entry, the FormatError and BundleError client
codes, the party_totals_left_out outcome and the saved-file schema,
takes party_totals: false as the request spelling of the cut, and is
recorded against dev at 133262c.
…303)

* fix(datasets): keep a year's whole totals and report each root's cut

The party totals cut decides each (time period, provenance) cell of an
open party role on its own. A cell where the role holds another party
loses its totals. A cell that holds only totals keeps the ones the code
index marks whole for that row and loses the rest; when none is whole,
it loses them all, unless the role holds no other party anywhere in the
frame, which keeps every total as before. So a year that publishes only
totals keeps one disjoint figure, and a cut that leaves nothing out
serves the frame's own version.

Trace entries gain an optional `kept_periods`, one entry per group of
totals kept in the same periods, and a stored trace replays exactly.
Each root of the derived manifest carries the `party_totals_left_out`
of its own version's trace. Ref roots are unchanged.

* fix(datasets): decide party totals within each series

The cut compared every series of a (time period, provenance) cell, so
one recipient's whole official-donor total dropped another recipient's
lone G7 row. It now decides each cell of a series: the frame's grain
less the role, time_period and provenance. A cell of only totals keeps
its whole totals; with none whole it drops them when the series holds
a member in another period and keeps them otherwise.

Trace entries replace `kept_periods` with `kept_in`, the cells of a
time period and series in which the listed totals stayed, so a stored
trace replays exactly.

* fix(datasets): replay recorded cuts and group kept cells by series

A recorded cut version re-rendered from its base frame, by the files
door or the refresh door, replays the cut its trace records, so it
re-hashes to its own content version under a changed rule or index.
A fresh mint still cuts with the current rule.

`kept_in` lists each series once with the time periods its totals
stayed in. The cut decides cells in Arrow per chunk, reads a role
without totals once, and drops rows with one semi-join per role and
dataset, so a frame with no totals cuts as fast as before.
…ts own table

A question naming its source ("from the OECD DAC2a") pins the part's provenance to that table, and `scope.provider._not_offered` refused the inclusive option for any pinned provenance, so France's net ODA to every recipient in 2023 offered only bilateral flows. The inclusive composite reads the table the bilateral answer reads, so a pin to that table loses nothing. `_not_offered` now refuses only a pin that admits another table, and `scope.edits._with_source` lets a stacked composite replace a pin to its bilateral component's table, which the CRS switch needed.
The client contract now names the frame party columns of each role
key. A contract test holds that table equal to ROLE_AXES.
…ive-pinned-provenance

fix(search): keep the inclusive provider option on a part pinned to its own table
The spec-kit pin comment named 133262c and d80df9f as the recording
commits; the client contract at 2f7462a is recorded at aacbe6a. The
contract test's docstring now lists the role-keys check.
…-kit-pins-sdk-0.2

chore(registry): pin spec-kit 2f7462a, the client contract for SDK 0.2
@jm-rivera
jm-rivera merged commit b6122d5 into main Nov 14, 2026
38 checks passed
@jm-rivera
jm-rivera deleted the release/2026-11-14-3 branch November 14, 2026 18:34

This branch was previously deployed

1 inactive deployment
dev — c9f3a0c5 Deployed Nov 14, 2026 by jm-rivera via deploy / deploy #188
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant